ci: extract cuda stage actions + runner_config mapping (#25138)

Co-authored-by: Alison Shao <54658187+alisonshao@users.noreply.github.com>
2026-05-13 21:16:57 -07:00
parent fd889097dc
commit 85d9c77c57
4 changed files with 478 additions and 826 deletions
@@ -0,0 +1,27 @@
+"""Emit a runner_config's setup details (install / artifact_version /
+install_timeout) in $GITHUB_OUTPUT format. Reads scripts/ci/runner_configs.yml.
+Called by .github/workflows/_pr-test-stage.yml.
+"""
+
+import os
+import sys
+
+import yaml
+
+_YAML_PATH = os.path.join(os.path.dirname(__file__), "runner_configs.yml")
+
+
+def load() -> dict:
+    with open(_YAML_PATH) as f:
+        return yaml.safe_load(f)["runner_configs"]
+
+
+if __name__ == "__main__":
+    if len(sys.argv) != 2:
+        sys.exit("usage: runner_configs.py <runner_config>")
+    rc = sys.argv[1]
+    config = load().get(rc)
+    if config is None:
+        sys.exit(f"unknown runner_config: {rc!r}")
+    for key, value in config.items():
+        print(f"{key}={value}")
@@ -0,0 +1,26 @@
+# Per-runner-config CUDA setup details. Single source of truth for the
+# `runner_config` field on `register_cuda_ci(...)` calls. Consumed by
+# scripts/ci/runner_configs.py (CLI wrapper), which is in turn called by
+# .github/workflows/_pr-test-stage.yml.
+#
+# Each runner_config carries install script, actions/download-artifact major
+# version, and install-step wall-clock cap (minutes, enforced via
+# `timeout-minutes:` on the install step in _pr-test-stage.yml).
+
+_anchors:
+  default_install: &default scripts/ci/cuda/ci_install_dependency.sh
+  deepep_install:  &deepep  scripts/ci/cuda/ci_install_deepep.sh
+  dsv4_install:    &dsv4    scripts/ci/cuda/ci_install_dsv4_dep.sh
+
+runner_configs:
+  1-gpu-small:       { install: *default, artifact_version: v4, install_timeout: "20" }
+  1-gpu-large:       { install: *default, artifact_version: v4, install_timeout: "20" }
+  2-gpu-large:       { install: *default, artifact_version: v4, install_timeout: "20" }
+  4-gpu-b200:        { install: *default, artifact_version: v6, install_timeout: "20" }
+  4-gpu-h100:        { install: *default, artifact_version: v4, install_timeout: "20" }
+  8-gpu-h200:        { install: *default, artifact_version: v4, install_timeout: "20" }
+  8-gpu-h20:         { install: *deepep,  artifact_version: v4, install_timeout: "20" }
+  deepep-4-gpu-h100: { install: *deepep,  artifact_version: v4, install_timeout: "20" }
+  deepep-8-gpu-h200: { install: *deepep,  artifact_version: v4, install_timeout: "20" }
+  dsv4-4-gpu-b200:   { install: *dsv4,    artifact_version: v6, install_timeout: "30" }
+  dsv4-8-gpu-h200:   { install: *dsv4,    artifact_version: v4, install_timeout: "30" }